GEOSCAN Search Results: Fastlink


TitleRetrieval of Surface Reflectance From Hyperspectral Data Using a Look-up Table Approach
AuthorStaenz, K; Williams, D
SourceCanadian Journal of Remote Sensing 23, 4, 1997 p. 354-368,
Alt SeriesEarth Sciences Sector, Contribution Series 20042370
PublisherInforma UK Limited
Mediapaper; on-line; digital
File formatpdf
Released2014 07 31
AbstractA five-dimensional look-up table (LUT) procedure has been implemented for the retrieval of surface reflectance from airborne and spaceborne hyperspectral data. This procedure can be applied in the visible and short-wave infrared up to 2700 nm. The dimensions of the LUT are: wavelength; pixel position; atmospheric water vapour content; aerosol optical depth; and terrain elevation. The LUTs are composed of simulated at-sensor radiance data generated by either the M5S (Modified 5S), LOWTRAN7, or MODTRAN2 or MODTRAN3 radiative transfer (RT) codes. Surface reflectances are computed by: convolving the LUTs data along the wavelength dimension with the spectral response profiles of the sensor bands; interpolating as necessary in the other four dimensions; fitting the data to a parametric equation. Additional features of the procedure include altitude-dependent layering of atmospheric aerosols and gases, scene-based estimation of atmospheric water vapour content, first order removal of adjacency effects, and choice of different exo-atmospheric solar irradiance functions. The LUT procedure can be applied to image cubes or individual spectra in forward (at-sensor radiance) and backward (surface reflectance) mode. Is is implemented in the Imaging Spectrometer Data Analysis System (ISDAS) which allows it to run with other tools that are relevant to surface reflectance retrieval, such as data pre-processing, post-processing, and visualization. Performance tests show good agreement between the LUT procedure and true RT code runs. Total relative errors of up to ±1.3% were achieved in case of MODTRAN2. These errors are a factor of about 0.4 lower than for M5S. Comparisons between reflectances retrieved from ground-based measurements and the LUT procedure indicate on average a relative agreement of better than 6% for the cases presented in this paper. This result was achieved using MODTRAN3. Although the computation time required to process an image cube can be quite long, the LUT procedure is substantially faster than running an RT code on a pixel-by-pixel basis. For instance, it takes about 15 hours on an ULTRASPARC 1 to process a full Airborne visible/Infrared Imaging Spectrometer (AVIRIS) cube of 614 pixels by 512 lines by 224 bands, using LUTs generated from 150 MODTRAN3 runs. This is about 3100 times faster that running this code on an individual pixel basis. The processing time for the same cube can be reduced to about 4 hours (130 times faster than individual RT code runs) if M5S is used instead of MODTRAN3. In this case, the accuracy is reduced, especially in the atmospheric absorption bands.

Date modified: